> ## Documentation Index > Fetch the complete documentation index at: https://docs.askfutures.com/llms.txt > Use this file to discover all available pages before exploring further. # Optimization > Sweep a strategy's parameters across ranges, let AskFutures test the grid, and find the best combination by the metric you care about — without overfitting. **Optimization** sweeps your strategy's tunable numbers across ranges, runs a backtest for every combination, and reports the one that scored best on the metric you chose. You pick the parameters and the ranges; AskFutures runs the grid and ranks the results. Every run in a sweep is a normal, deterministic backtest — same real prices, same modeled costs. Optimization just runs many of them and compares the numbers. The AI doesn't decide what's "best"; the simulator does the math and the metric does the ranking. ## How a sweep works Any tunable number on the strategy card — an EMA period, a stop size, a target, an ATR multiple. Hold the rest fixed. Give a **min**, a **max**, and either a **step** or a number of test points. For example, sweep the stop from $200 to $600 in \$100 steps (5 values), or the fast EMA from 5 to 15. Best average P\&L per trade, best total P\&L, best win rate — whatever defines "better" for your idea. AskFutures backtests every combination in the grid and reports the winning parameter set, with the full table available to download. The strategy card shows the size of the job before you commit — something like *"1 parameter · 10 runs"* — and you can switch each range between an evenly spaced count (**Num. steps**) or your own explicit values (**Custom**), on a **linear** or **log** scale. Two ranges multiply: a 5-value stop sweep crossed with a 4-value target sweep is 20 backtests. ## Two-tier sweeps: why some changes are fast and some are slow Under the hood, a sweep is organized in two tiers — an **outer** loop and an **inner** loop — and which tier a parameter lands in is why some sweeps fly and others take a while. ```mermaid theme={null} flowchart LR O["Outer loop
changes the indicators
(recompute everything)"] --> I["Inner loop
changes only the trade rules
(reuse the indicators)"] I --> R["Score & rank
by your metric"] ``` Parameters that only affect **how trades are managed** — stop size, target, trailing stop, max time in trade. The indicators don't change, so the engine reuses them and just re-simulates the exits. Many values, cheaply. Parameters that change an **indicator itself** — an EMA period, an RSI length, an ATR window. Every value forces the indicators to be recomputed across the whole window before the trades can run. Fewer values, more work each. Want a big sweep without a long wait? Put your wide ranges on the **inner** parameters (stops and targets) and keep the **outer** parameters (indicator lengths) to a handful of values. ## When to optimize Optimize **after** a strategy already backtests cleanly and shows a plausible edge — to refine it, not to rescue it. The strategy passes its backtest, takes a sensible number of trades, and the average P\&L per trade is around break-even or better. Now you're asking *"is there a better stop, or a slightly different EMA pair?"* — a refinement. The strategy fails to run, takes a handful of trades, or loses badly across the board. No range of stop sizes fixes a broken idea — fix the logic in chat first, then sweep. ## The overfitting trap **A sweep finds the parameters that fit the *past* best — which is not the same as the parameters that will work next.** The more combinations you try, the more likely the "winner" is just the one that happened to line up with old noise. Treat the best result as a hypothesis, not a guarantee. Guard against it: * Prefer **wide, robust plateaus** over a single razor-thin peak. If only one exact value works and its neighbors fall apart, that's a red flag. * Keep sweeps **small and meaningful** — a few parameters, sensible ranges. * Re-test the winner on a **different window** (e.g. a more recent stretch you didn't sweep over) before you trust it. Past performance does not guarantee future results. Always test before you trade. For the full picture — why the top-ranked result is often the least trustworthy one, and how to read a sweep as a robustness map — see [overfitting vs. optimization](/concepts/overfitting-vs-optimization). ## Reading the results Each sweep finishes as a labelled batch — e.g. *"Optimization 1 · Completed · 6 combinations · Best: \$-2.97 avg"* — naming the winning parameters and the score on your chosen metric. Open it to see the ranking, or download the full table as a CSV to compare every combination yourself. Found a winner you like? Ask AskFutures to apply it. The chosen parameters become a new saved version of the strategy, so you can [compare](/concepts/strategies) it against the original. ## Next steps